Starlink Mobile Comes Down to Earth: What the Qualcomm Deal Really Tells Us

SpaceX is no longer only a satellite company that partners with mobile operators. With today’s Qualcomm deal, it is becoming a mobile operator in its own right, with real base stations on real towers.

I have spent more than twenty years in this industry, first on the vendor side with Huawei and Nokia, then on the operator side at Orange. From both sides of the table, this announcement reads less like a chip contract and more like a declaration of intent. Starlink Mobile wants to compete with AT&T, Verizon and T-Mobile on their own ground, and it has now chosen the silicon to do it.

What was announced

On 4 November 2026, Qualcomm confirmed a multi-year agreement to supply the radio, baseband and software for Starlink Mobile’s nationwide terrestrial 5G network in the United States, with deployment starting in 2027.

This silicon is not experimental. The same platforms already run in commercial networks, including NTT Docomo’s nationwide 5G vRAN in Japan and Viettel in Vietnam.

It also closes a few months of speculation. Back in August, Elon Musk floated the idea of small femtocells mounted next to Starlink dishes, and the market then wondered whether SpaceX would simply buy a US carrier. Instead, SpaceX has chosen the classic operator path: macro sites with high-end radios.

Why go terrestrial: physics always wins

Direct-to-cell from low Earth orbit is a brilliant coverage layer, but it is not a capacity layer, and that is the whole reason for this move.

A satellite beam covers a very large area on the ground, so the capacity of that beam is shared by every phone underneath it. The phone itself is the other limit, because a normal smartphone has a tiny antenna and a few hundred milliwatts of power, while the satellite sits several hundred kilometres away. In practice, that gives you texting, basic data and voice in rural areas, which is valuable, but it cannot carry a city at rush hour or deliver the indoor experience people expect from their carrier.

Anyone who has planned a network for a large event knows the answer: you add sites, you add antennas, and you bring capacity close to the user. SpaceX has reached the same conclusion. To be a real carrier, it needs towers on the ground and satellites in the sky, working as one network.

Reading the RAN design like an architect

The most interesting detail is not the chip name. It is the choice of 64T64R Massive MIMO on bands below 2.5 GHz, which is an unusual and ambitious combination.

In most networks, 64T64R radios are used on 3.5 GHz, where the wavelength is short and you can pack 64 antenna elements into a panel that a tower can carry. At 1.7 or 2.1 GHz, the wavelength is roughly twice as long, so the same array becomes much larger and heavier. This is why Qualcomm talks about new beamforming work for lower bands, because making massive MIMO practical at mid-low band is a real engineering problem, not a marketing line.

The reward, if they solve it, is large. Lower bands travel further and go through walls better, and massive MIMO multiplies capacity through beamforming and multi-user MIMO. In simple terms, SpaceX is trying to get the coverage of a low band and the capacity of a C-band site from the same spectrum, which is exactly what an operator with limited spectrum needs.

The split between RU and DU also matters. The RU handles the radio and the lowest part of the physical layer, while the DU handles the real-time Layer 1 and Layer 2 processing and serves several radios. With the QDU100 doing Layer 1 on dedicated silicon, the DU can run as software on standard servers, which is the virtualised RAN model that Docomo proved at national scale.

Open RAN’s second chance, and the DISH lesson

This deal gives Open RAN its most credible greenfield test since DISH, and with some irony it uses much of the same spectrum DISH once held through EchoStar.

The QRU100 is designed as an O-RAN compliant platform, so the direction is clear even if SpaceX has not yet named its full vendor list. A disaggregated RAN lets SpaceX pick radios, servers, software and the core separately, and run the whole network with its own automation. For a company that writes its own software for rockets and satellites, that control is the real attraction.

But DISH taught the industry a hard lesson. The technology worked, yet integration took longer than planned, site rollout was slow, and the commercial base never grew fast enough to pay for the network. Open RAN does not remove the integration work; it moves it from the vendor to the operator. SpaceX has two things DISH lacked: a very strong engineering culture and an existing customer base of millions of Starlink users who can be offered a mobile plan. Whether that is enough is the big question for 2027 and 2028.

The real play: one network, two layers

The strategic value is not the terrestrial network alone. It is a single operator that owns both the ground layer and the space layer, under one core, one SIM and one brand.

Today, every satellite-to-phone service depends on a partner carrier, its spectrum and its roaming agreements. Starlink Mobile could remove that dependency. Towers would carry the traffic in cities and suburbs, while satellites would fill every gap in between, from highways to national parks to the ocean. With 3GPP NR-NTN now in the standards, both layers can, in principle, be managed by the same core network and handled by the same chipset in the phone.

No US carrier can offer that today without a partner. That is why this announcement matters more than the size of the first rollout.

Who should pay attention

For operators in Africa and the Middle East, the last row is the one to watch. Our regulators are already reviewing satellite broadband and direct-to-device licences. The question is no longer whether satellite players will offer mobile services, but whether they will partner with local operators or try to become operators themselves.

The hard questions nobody has answered yet

Choosing a chip is the easy part of building a mobile network. The difficult parts are slow, local and expensive, and they will decide whether Starlink Mobile succeeds.

• Sites: a national network needs tens of thousands of locations, each with a lease, a permit and power. Site acquisition, not technology, is usually what slows a rollout.
• Transport: every site needs backhaul. Fibre is costly, and satellite backhaul adds latency, so the transport plan will shape both cost and quality.
• Spectrum depth: the AWS/MSS holdings are useful but limited. A nationwide carrier usually needs a low band for coverage and more mid-band for capacity, which suggests more spectrum deals ahead.
• Core and devices: SpaceX has not detailed its 5G core, its voice solution (VoNR and IMS), or its roaming strategy. Device support for the exact band plan will also take time.
• Capex: a nationwide network costs tens of billions of dollars. SpaceX has deep pockets, but the payback depends on how many Starlink users convert to mobile plans.

My take

SpaceX has stopped asking whether it needs the carriers and has started building what the carriers have. The Qualcomm deal is the first concrete step from a satellite service toward a full hybrid operator.

The technology choices are sound: proven silicon, an open architecture, and an ambitious bet on massive MIMO at lower frequencies. The real test will come on the ground, in site permits, backhaul, spectrum and execution speed, where many greenfield operators have struggled before. If SpaceX gets those right, 2027 could be the year the line between terrestrial and non-terrestrial networks starts to disappear.

I would like to hear from operators and engineers: do you see Starlink Mobile as a serious fourth carrier, or as another DISH in the making? Share your view in the comments.